Prediction of heat transfer parameters of nuclear reactor based on physical information machine learning algorithm

The accurate prediction of the heat transfer coefficient (HTC) under extremely high parameter conditions in nuclear reactors is crucial for reactor design and operation. However, this involves complex phenomena influenced by multiple factors across different flow patterns, and the physical mechanisms are not fully understood. Due to the lack of experimental data that meet the high temperature and pressure parameters of actual reactors, semi-empirical relationships that heavily rely on experimental data find it difficult to meet the requirements for high-precision numerical calculations in nuclear reactors.Deep learning algorithms can effectively solve complex nonlinear problems, but they have limitations such as poor extrapolation performance and overfitting. This study develops an HTC prediction model by combining prior physical information from the Jens-Lottes relationship and Thom relationship with machine learning algorithms, including multilayer perceptrons, backpropagation neural networks, and random forests. The model is trained and validated using experimental data from circular tube channels, and the applicability and prediction accuracy of six different machine learning algorithm models based on physical information are evaluated.